Physx on AMD CARDS!!!!!????

R

ratnadeep007gamer

Broken In
I have a very different type of question in my mind.
We all know that nvidia has Physx in their cards to unload the pressure of physx calculation from the CPU and let it done by the GPU. And also that physx is technology by Ageia and it was bought by nvidia and this is fairly proprietary on their cards by their drivers. Does AMD by now had developed something similar to physx to unload the pressure of physx from CPU to their GPU via any driver.

Is their any third party driver to do the same.

Is not having physx does lower the FPS in the games running on AMD GPU in same CPU and system specs.

And What is 'CUDA' as nvidia very proudly writes 'CUDA powered' and similarly had AMD having similar thing and does not having that decrease performance.

I prefer nvidia because of only the above fact,i.e., Physx. Am I wrong?

I was comparing HD Radeon graphics with I have a very different type of question in my mind.
We all know that nvidia has Physx in their cards to unload the pressure of physx calculation from the CPU and let it done by the GPU. And also that physx is technology by Ageia and it was bought by nvidia and this is fairly proprietary on their cards by their drivers. Does AMD by now had developed something similar to physx to unload the pressure of physx from CPU to their GPU via any driver.

Is their any third party driver to do the same.

Is not having physx does lower the FPS in the games running on AMD GPU in same CPU and system specs.

And What is 'CUDA' as nvidia very proudly writes 'CUDA powered' and similarly had AMD having similar thing and does not having that decrease performance.

I prefer nvidia because of only the above fact,i.e., Physx. Am I wrong?

I was comaparing GT650 withI have a very different type of question in my mind.
We all know that nvidia has Physx in their cards to unload the pressure of physx calculation from the CPU and let it done by the GPU. And also that physx is technology by Ageia and it was bought by nvidia and this is fairly proprietary on their cards by their drivers. Does AMD by now had developed something similar to physx to unload the pressure of physx from CPU to their GPU via any driver.

Is their any third party driver to do the same.

Is not having physx does lower the FPS in the games running on AMD GPU in same CPU and system specs.

And What is 'CUDA' as nvidia very proudly writes 'CUDA powered' and similarly had AMD having similar thing and does not having that decrease performance.

I prefer nvidia because of only the above fact,i.e., Physx. Am I wrong?

I was comaparing HD Radeon Graphics with nVIDIA Graphics having same clock speed and same other specs etc
 
C

Coldbreeze16

Journeyman
IMHO it doesn't affect any game's playability that much. Games, by themselves don't fully stress the CPU so it isn't that relevant to consider offloading the work to GPU. Besides GPU based physx is supported by a few titles: In most cases you won't even notice that extra PhysX is there (my opinion is only Batman:AA is one exception, could be wrong). And give it a year and two and the CPU's will catch up. So let PhysX not be a major deciding feature.
List of games with hardware-accelerated PhysX support - Wikipedia, the free encyclopedia

CUDA core is nVidia's fancy name for what AMD calls stream processors. I think nVidia also called them Stream Processors earlier. But just like AMD and Intel's architecture, they are also some distance apart. AMD's answer to everything is 'stuff up more cores' :lol: Their individual SP's lack a bit of power vs. CUDA but AMD makes up for it with extra cores. Plus at the same price AMD cards give much more headroom for OC which can, in most cases, bring it up to next tier nVidia cards. many AMD cards also have extra memory bandwidth compared to similarly priced nVidia offerings. AMD is real VFM.

If you post exactly which cards you are comparing or what your budget is, we could make a suggestion.
 
Cilus

Cilus

laborare est orare
PhysX is proprietary things to nVidia and runs either on CPU or only in nVidia GPU. It does not run on any other Graphics cards apart from nVidia. And CUDA is not just the fancy name of the nVidia stream processors, it's a programming language for GP-GPU computing in nvidia Hardware. Initially PhysX used to run very poorly on any x86 processors because the CPU code path of PhysX was based on very old x87 code which is very poorly optimized for modern CPU whereas code path for the nVidia GPU was written in CUDA language which is far optimized.

However, latest PhysX SDK (PhysX SDK 3.XX ) now providing x86 CPU optimization and can use advanced instruction sets like SSE2 and SSE3 and games implementing those features will run efficiently on any x86 CPU apart from nVidia GPU. So in future titles with implementing PhsyX with PhysX 3.xx SDK, expect smooth performance by using CPU if you don't have a nVidia GPU.

CUDA core is nVidia's fancy name for what AMD calls stream processors. I think nVidia also called them Stream Processors earlier. But just like AMD and Intel's architecture, they are also some distance apart. AMD's answer to everything is 'stuff up more cores' Their individual SP's lack a bit of power vs. CUDA but AMD makes up for it with extra cores. Plus at the same price AMD cards give much more headroom for OC which can, in most cases, bring it up to next tier nVidia cards. many AMD cards also have extra memory bandwidth compared to similarly priced nVidia offerings. AMD is real VFM.
Click to expand...

Where did you get this? Do you have an understanding of AMD's current GCN or Graphics Core Next architecture which is powering the HD 7000 series cards?
 
harshilsharma63

harshilsharma63

DIY FTW!
Coldbreeze16 said:
IMHO it doesn't affect any game's playability that much. Games, by themselves don't fully stress the CPU so it isn't that relevant to consider offloading the work to GPU. Besides GPU based physx is supported by a few titles: In most cases you won't even notice that extra PhysX is there (my opinion is only Batman:AA is one exception, could be wrong). And give it a year and two and the CPU's will catch up. So let PhysX not be a major deciding feature.
List of games with hardware-accelerated PhysX support - Wikipedia, the free encyclopedia

CUDA core is nVidia's fancy name for what AMD calls stream processors. I think nVidia also called them Stream Processors earlier. But just like AMD and Intel's architecture, they are also some distance apart. AMD's answer to everything is 'stuff up more cores' :lol: Their individual SP's lack a bit of power vs. CUDA but AMD makes up for it with extra cores. Plus at the same price AMD cards give much more headroom for OC which can, in most cases, bring it up to next tier nVidia cards. many AMD cards also have extra memory bandwidth compared to similarly priced nVidia offerings. AMD is real VFM.

If you post exactly which cards you are comparing or what your budget is, we could make a suggestion.
Click to expand...

:facepalm:

from CUDA programming guide by nvidia (CUDA C Programming Guide :: CUDA Toolkit Documentation)

The NVIDIA GPU architecture is built around a scalable array of multithreaded Streaming Multiprocessors (SMs). When a CUDA program on the host CPU invokes a kernel grid, the blocks of the grid are enumerated and distributed to multiprocessors with available execution capacity. The threads of a thread block execute concurrently on one multiprocessor, and multiple thread blocks can execute concurrently on one multiprocessor. As thread blocks terminate, new blocks are launched on the vacated multiprocessors.
Click to expand...

@coldbreeze; please dont post something you don't know about.

*en.wikipedia.org/wiki/Stream_processing
 
Last edited:
OP
R

ratnadeep007gamer

Broken In
I was comparing the following graphics cards for my friends laptops.
nvidia cards:
GT650M,GT730M,GT740M
radeon cards
7660G+7670M,7970M,7660M(alone),8730M,8750M

Thanks for clearing CUDA with the link
CUDA C Programming Guide :: CUDA Toolkit Documentation
 
C

Coldbreeze16

Journeyman
They are different architectures but still accomplish same things, kindly enlighten me with a link if I'm wrong. Didn't nVidia call them SP before the Fermi architecture? I mean similar and not same. They're still the frigging shader units. Of course I said technically CUDA core = SP and not CUDA = SP (CUDA= ATI Stream). Whether nVidia has proprietary CUDA SDK apart from OpenCL or not: that's something a gamer won't need to dabble with.

I haven't read up on GCN but despite having 2-3 times SP compared to CUDA cores on similarly priced nVidia cards and seeing what AMD does on the processor front, I surmised that. And I have googled it earlier, most discussions tend to favor that hypothesis. Please link me if there's evidence to the contrary. I googled again and here's one of many that I found What's better cuda core or stream processors? - Yahoo! Answers

Addendum: Googled again and I see nVidia called them processing cores and not SP.

*www.luxrender.net/forum/viewtopic.php?f=13&t=4384

*superuser.com/questions/331591/what-are-radeon-stream-processors-and-simd-engines
 
Last edited:
Cilus

Cilus

laborare est orare
Stream Processor is a term for each of the processing elements in a SIMD (single Instruction multiple Data) unit which is cluster of those execution units. But they are not at all multiprocessor unit. And after the adaptation shader architecture, starting from from nVidia's 8000 series and AMD's 3000 series, each of the execution units of a GPU is called Stream Processor or SP. each of the new architecture release from both the camp have improved their working capabilities, speed, power efficiency; but they are still called stream processors.

Regarding ATI Stream, that has been dropped log back and now they are calling it APP which is mainly based on OpenCL implementation instead of any proprietary techniques.

Coldbreeze16, you really don't have the knowledge of a vector processor architecture and therefore not understanding a single thing here. Just googling is not going to help you. Try download and read the official articles of GCN and nvidia Kepler, released by their respective manufacturers.
I haven't read up on GCN but despite having 2-3 times SP compared to CUDA cores on similarly priced nVidia cards and seeing what AMD does on the processor front, I surmised that.
Click to expand...

A HD 7970 GHz edition does have 2048 Stream Processors of GCN architecture and a GTX 680 does have 1536 CUDA cores based on Kepler architecture and the GHz edition performs better than a GTX 680, especially in 2560X1600 or XHD resolution. Now I don't think neither 1536 X 3 = 2048 nor 1536 X 2 = 2048. So get your facts current before posting. Otherwise it is considered as gibberish. ANd please don't compare a x86 processor architecture with a SIMD design which relies on Vector Processor design
 
C

Coldbreeze16

Journeyman
Cilus said:
Stream Processor is a term for each of the processing elements in a SIMD (single Instruction multiple Data) unit which is cluster of those execution units. But they are not at all multiprocessor unit. And after the adaptation shader architecture, starting from from nVidia's 8000 series and AMD's 3000 series, each of the execution units of a GPU is called Stream Processor or SP. each of the new architecture release from both the camp have improved their working capabilities, speed, power efficiency; but they are still called stream processors.

Regarding ATI Stream, that has been dropped log back and now they are calling it APP which is mainly based on OpenCL implementation instead of any proprietary techniques.

Coldbreeze16, you really don't have the knowledge of a vector processor architecture and therefore not understanding a single thing here. Just googling is not going to help you. Try download and read the official articles of GCN and nvidia Kepler, released by their respective manufacturers.


A HD 7970 GHz edition does have 2048 Stream Processors of GCN architecture and a GTX 680 does have 1536 CUDA cores based on Kepler architecture and the GHz edition performs better than a GTX 680, especially in 2560X1600 or XHD resolution. Now I don't think neither 1536 X 3 = 2048 nor 1536 X 2 = 2048. So get your facts current before posting. Otherwise it is considered as gibberish. ANd please don't compare a x86 processor architecture with a SIMD design which relies on Vector Processor design
Click to expand...

You're correct, my assumptions were based on older VLIW. I downloaded a presentation on GCN.

But this will be like splitting hair, all I mentioned was for practical purposes both SP are CUDA cores are shader units.
 
Cilus

Cilus

laborare est orare
^^ Yes both SP and CUDA cores are sharder units and you're right about it. They care called CUDA core becayse nVidia's CUDA language targets them to exeute applications in parallel.
\And Regarding VLIW, don't get confused with the high number of Shader units in older (HD 4000, HD 5000 and HD 6000 series) Radeon cards. In VLIW5 or in VLIW4, 5 execution units or 4 units make a single thread processing units. All those 5 (in VLIW5) or 4 (in VLIW4) can't operate on different threads, instead they work on different instructions of a single thread. So, for a GPU having 1000 VLIW5 SP means it actually has (1000/5)= 200, 5 dimentional effective SP. VLIW is very much suitable for Graphics processing and static programming but not so good for Compute tasks.
 
OP
R

ratnadeep007gamer

Broken In
So, is it good to go with AMD or should I stick with nvidia. And how to know before buying card that is it a 64-bit bandwidth or 128-bit because I saw that nvidia a has same cards but having different bandwidth. And how can I know that in laptops is it a 64-bit or 128-bit and does bandwidth decreasing affect the performancE.

As i found in Lenovo Z500(59-380480) and Lenovo Z500(59-380463) they has GT-740M 64-bit but memory speed and clock speed is greater than the 128-bit. So, how can I know that company is giving me 64-bit or 128-bit specially in lappy before buying.

So, is it good to go with AMD or should I stick with nvidia. And how to know before buying card that is it a 64-bit bandwidth or 128-bit because I saw that nvidia a has same cards but having different bandwidth. And how can I know that in laptops is it a 64-bit or 128-bit and does bandwidth decreasing affect the performancE.

As i found in Lenovo Z500(59-380480) and Lenovo Z500(59-380463) they has GT-740M 64-bit but memory speed and clock speed is greater than the 128-bit. So, how can I know that company is giving me 64-bit or 128-bit specially in lappy before buying.
 
Cilus

Cilus

laborare est orare
Nvidia's laptop or mobility card series are real buggy and they have different versions of a card but different hardware configuration and identifying which laptop has which version of card is real pain in ass. There is no easy way to do that you need to search different forums and Notebook review sites to find the Laptop model you have chosen and the details about it.
 
C

Coldbreeze16

Journeyman
You actually want to buy a graphics card or just want to judge which camp is better? Since you mentioned GT650 by which you mean 650M. How do you suppose you will change the GPU in a laptop? :s AFAIK That's on the verge of impossible.
 
OP
R

ratnadeep007gamer

Broken In
Coldbreeze16 said:
You actually want to buy a graphics card or just want to judge which camp is better? Since you mentioned GT650 by which you mean 650M. How do you suppose you will change the GPU in a laptop? :s AFAIK That's on the verge of impossible.
Click to expand...

I don't want to change gpu in laptops just gaining knowledge and comparing the things. And which will play games easily.


I have one more question what are general purpose gpu (GPGPU).

Coldbreeze16 said:
You actually want to buy a graphics card or just want to judge which camp is better? Since you mentioned GT650 by which you mean 650M. How do you suppose you will change the GPU in a laptop? :s AFAIK That's on the verge of impossible.
Click to expand...

I don't want to change gpu in laptops just gaining knowledge and comparing the things. And which will play games easily.


I have one more question what are general purpose gpu (GPGPU).
 
C

Coldbreeze16

Journeyman
General-purpose computing on graphics processing units - Wikipedia, the free encyclopedia

The name should be self explanatory, a graphics card has a proccy that handles graphics processing and its instuction set is optimized for that mostly. The newer GPU's however can also do General Purpose (GP) computing traditionally handled by CPU.
 
B

bhvm

Journeyman
I remember there was a separate "AGEIA PHYSX" PCI-e card until Nividia buyed off Ageia. It worked only 1 or 2 games back then. Ghost recon Advanced warfighter 2.

Poor lads who brought the card.
 
You must log in or register to reply here.
Top Bottom